System and method for medical coding of vascular interventional radiology procedures

Abstract

A system and method for identifying medical procedure codes and medical diagnosis codes from physician reports that describe a vascular interventional radiology procedure using a combination of natural language processing (NLP) and human medical coders. In one embodiment, the system and method of the present invention creates billing results, and or other documents, that are compliant with applicable legal and policy instructions from the government or a medical institution. Medical billing codes are efficiently extracted from medical reports using a. NLP engine and a graphical user interface optimized for understanding the VasIR medical procedure described in the report.

What is claimed: 
     
         1 . A method comprising:
 receiving, from a remote system via a network connection, a first medical report containing natural language;   statistically analyzing, using a computer, a set of training medical reports having pre-annotated sentence elements, wherein each sentence element is a sentence or a phrase;   calculating, using the computer, a conditional probability for each sentence element of the set of training medical reports, the conditional probability indicating the probability that the sentence element belongs to a specific functional region in the set of training medical reports;   calculating, using the computer and the conditional probability for each sentence element of the set of training medical reports, a run-time probability for each sentence element of the first medical report, the rim-time probability indicating the probability that the sentence element belongs to a specific functional region in the first medical report; and   assigning, using the computer and the run-time probability, each sentence element of the first medical report to the corresponding specific functional region in the first medical report.   
     
     
         2 . The method of  claim 1 , further comprising:
 extracting, by the computer, at least one language-based feature from the functional regions of the first medical report;   identifying, via the computer, at least one standardized code associated with the extracted at least one language-based feature;   generating, via the computer, a confidence assessment for the at least one standardized code, wherein the generated confidence assessment indicates the likelihood that the identified standardized code is correct given evidence including the at least one extracted language-based feature; and   generating a medical report-level confidence assessment based on the confidence assessment associated with the at least one standardized code.   
     
     
         3 . The method of  claim 1 , wherein the first medical report comprises a set of medical reports. 
     
     
         4 . The method of  claim 1 , wherein the first medical report comprises information relating to one or more medical reports, the information being obtained from multiple remote data feeds and merged to create the first medical report. 
     
     
         5 . The method of  claim 1 , where the step of calculating the run-time probability for each sentence element of the first medical report uses a Markov model. 
     
     
         6 . The method of  claim 2 , further comprising routing the first medical report to one of a plurality of queues based on the report-level confidence assessment. 
     
     
         7 . The method of  claim 1 , further comprising calculating a regional transition probability between adjacent sentence elements in the set of training, medical reports, and using the regional transition probability in the step of calculating the run-time probability for each sentence element of the first medical report. 
     
     
         8 . A system comprising:
 one or more first computers, each of the first computers being configured perform a regioning process, the regioning process assigning each sentence element of a first medical report containing natural language to a corresponding functional region in the first medical report by:   receiving, by the first computers from a remote system via a network connection, the first medical report;   statistically analyzing, using the first computers, a set of training medical reports having pre-annotated sentence elements, wherein each sentence element is a sentence or a phrase;   calculating, using the first computers, a conditional probability for each sentence element of the set of training medical reports, the conditional probability indicating the probability that the sentence element belongs to a specific functional region in the set of training medical reports;   calculating, using the first computers and the conditional probability for each sentence element of the set of training medical reports, a run-time probability for each sentence element of the first medical report, the run-time probability indicating the probability that the sentence element belongs to a specific functional region in the first medical report; and   assigning, using the computer and the run-time probability, each sentence element of the first, medical report to the corresponding specific functional region in the first medical report.   
     
     
         9 . The system of  claim 8 , wherein each of the first computers is further configured to identify a set of one or more standardized codes and generate a confidence assessment for the identified set of standardized codes by:
 extracting, by the computer, at least one language-based feature from the functional regions of the first medical report;   identifying, via the computer, one or more standardized codes associated with the extracted at least one language-based feature; and   generating, via the computer, a confidence assessment for the one or more standardized codes, wherein the generated confidence assessment indicates the likelihood that the one or more standardized codes are correct given evidence including the at least one extracted language-based feature.   
     
     
         10 . The system of  claim 8 , wherein the first medical report comprises a set of medical reports. 
     
     
         11 . The system of  claim 8 , wherein the first medical report comprises information relating to one or more medical reports, the information being obtained from multiple remote data feeds and merged to create the first medical report. 
     
     
         12 . The system of  claim 8 , where the step of calculating the run-time probability for each sentence element of the first medical report uses a Markov model. 
     
     
         13 . The system of  claim 9 , further comprising a second computer, the second computer configured to perform a learning process by:
 receiving the one or more standardized codes identified by the one or more first computers and the confidence assessment generated by the one or more first computers for the one or more standardized codes;   displaying, via a graphical user interface, the one or more standardized codes to a human coder;   receiving, for the one or more standardized codes, an indication as to whether a human coder approved or modified the one or more standardized codes;   adjusting the confidence assessment generating step based on the received indications as to whether a human coder approved or modified the one or more standardized codes selecting one of a plurality of queues based on the generated confidence assessment; and   routing the medical report to the selected queue.   
     
     
         14 . The system of  claim 13 , wherein the second computer is different from any of the one or more first computers. 
     
     
         15 . The system of  claim 8 , wherein each of the first computers is further configured to calculate a regional transition probability between adjacent sentence elements in the set of training medical reports, and use the regional transition probability in the step of calculating the run-time probability for each sentence element of the first medical report.